1: \begin{abstract}

2: In this article, we consider multilevel Monte Carlo for the numerical

3: computation

4: of expectations for stochastic differential equations driven by L\'evy

5: processes. The underlying numerical schemes are based on jump-adapted

6: Euler schemes. We prove stable convergence of an idealised scheme.

7: Further, we deduce limit theorems for certain classes of functionals

8: depending on the whole trajectory of the process. In particular, we

9: allow dependence on marginals, integral averages and the supremum of

10: the process. The idealised scheme is related to two practically

11: implementable schemes and corresponding\vspace*{1pt} central limit theorems are

12: given. In all cases, we obtain errors of order $N^{-1/2} (\log

13: N)^{1/2}$ in the computational time $N$ which is the same order as

14: obtained in the classical set-up analysed by Giles~[\textit{Oper. Res.} \textbf{56} (2008) 607--617].

15: Finally, we use the central limit theorems to optimise the parameters

16: of the multilevel scheme.

17: \end{abstract}